Role: Sr Staff TLM, MultiModal AI
Computer vision work for Flock Safety, whose camera network helps law enforcement investigate crime across thousands of US communities. Two products in particular:
Flock National LPR Network
A nationwide license-plate-reader network that processes more than 20 billion plate reads per month across 5,000+ communities and 4,800+ agencies. The vision pipeline detects and matches plates in real time, flags hits against shared hotlists and NCIC/FBI databases, and pushes actionable alerts to officers in the field. Product page.
Flock FreeForm
A natural-language search layer over video and LPR footage. Instead of rigid filter menus, an investigator can type something like “man in a blue shirt and cowboy hat” and the system retrieves matching frames across their own cameras and partner agencies' feeds, with safeguards to keep queries from biased searches. Product page.
Technical Highlights:
- Plate detection and OCR tuned for low-light, oblique angles, and partially occluded plates across hundreds of camera models.
- Multimodal retrieval — image-text embeddings over video frames — so free-text descriptions surface the right clips in seconds.
- Scaled inference to keep up with 20B+ monthly reads while holding a tight false-positive budget for downstream alerts.
